Enumeration of Plasmodium Parasites on Thin Blood Smear Digital Microscopic Images

An automated parasite detection is one of the challenges in computer vision and medicine field. Recently, an automated detection system became necessary in order to assist the parasitologist in detecting parasite. However, the development of automated parasite detection had several challenges. Mostly, previous research studies were conducted with the homogeny dataset. Hence, it will become a problem if the research was conducted with various dataset. According to those facts, this study proposed a scheme for detecting Plasmodium parasites. In this study, 46 thin blood smear digital microscopic images of Plasmodium Falciparum with several kinds of pixel intensity were used. The proposed method started by normalizing the data. The normalization result was then segmented by using Otsu thresholding and some operations of morphological operation. Based on the segmentation result, parasite candidate detection was done by storing the location of an object in segmentation result and labeling the object by using BLOB analysis. The result of detection process produced high number of false positive. To reduce the number of false positive, feature extraction process followed by classification process was conducted. Several texture features such as histogram, GLCM and GLRLM were used to extract information of parasite. Classification process was conducted by applying several kernels of SVM classifier. According to the evaluation process, the proposed method has better performance than previous study and successfully distinguish parasite and non-parasite by using the cubic kernel of SVM with accuracy, sensitivity, specificity, PPV and NPV of 97.4%, 100%, 94.8%, 95.1% and 100%, respectively. These results indicated that the proposed scheme is proper in order to assist the expert in detecting Plasmodium parasites effectively.